OBJECTIVES: Estimation of cost differences between groups has been assessed using a variety of econometric techniques. Recent literature commonly uses ordinary least squares (OLS) with log transformed costs or generalized linear models with a log link function and gamma error distribution; however recommendations from the field of econometrics suggest that quantile regression models may provide more informative comparisons of healthcare cost data. The objective of this study was to compare three econometric models for estimating healthcare cost differences. METHODS: The econometric models were compared using a cross-sectional sample of patients hospitalized during 2007 for COPD exacerbations from Premier’s Perspective Comparative Database, a population-based U.S. inpatient database. Admissions were classified into two groups according to the presence of pneumonia. Total inpatient cost was regressed on pneumonia status and confounders using: 1) OLS with log-transformed costs; 2) generalized linear models with log link and gamma error; and 3) quantile regression with comparisons at every 5th percentile. Cost differences were calculated by subtracting model based estimated costs between COPD patients with and without pneumonia. RESULTS: There were 69,841 COPD exacerbation admissions. Of these, 69,286 admissions had nonzero costs, and 6,840 (9.9%) had evidence of pneumonia. In the OLS model, estimated costs were $3,949 (95%CI: $3,822 - $4,076) higher in patients with pneumonia; in the log-gamma model, the estimated costs were $4,442 (95%CI: $4,297 - $4,587) higher; while the median quantile difference was $3,756 (95%CI: $3,636 - $3,876).  The quantile regression differences ranged from a low of $2,078 (95%CI: $2,012 - $2,145) for the 5th quantile to a high of $8,691 (95%CI: $8,395 - $8,987) for the 95th quantile. CONCLUSIONS: Results from econometric regression models can vary greatly depending on model assumptions. Quantile regression results provide a detailed picture of cost differences across the entire cost distribution, illustrating important non-uniformities of differences between groups.